Synthetic flight performance data generation
Systems and methods of synthetic flight performance data generation include obtaining training data corresponding to multiple timeseries of multivariate aircraft performance data of actual aircraft flights. The systems and methods also include performing a training operation of a generative adversarial network that includes a synthetic aircraft performance data generator and a discriminator. After completion of the training operation, the systems and methods also include receiving one or more input parameters, and generating, at the synthetic aircraft performance data generator, one or more timeseries of synthetic aircraft performance data based on the one or more input parameters.
1 . A method comprising:
obtaining training data corresponding to multiple timeseries of multivariate aircraft performance data of actual aircraft flights;
processing the training data to remove outlier values and add missing values;
performing a training operation of a generative adversarial network that includes a synthetic aircraft performance data generator and a discriminator, the training operation including:
updating the synthetic aircraft performance data generator to improve a likelihood that synthetic aircraft performance data generated by the synthetic aircraft performance data generator is identified as actual aircraft performance data by the discriminator; and
updating the discriminator to improve an ability of the discriminator to distinguish between the training data and the synthetic aircraft performance data generated by the synthetic aircraft performance data generator; and
after completion of the training operation:
receiving one or more input parameters; and
generating, at the synthetic aircraft performance data generator, one or more timeseries of synthetic aircraft performance data based on the one or more input parameters.
2 . The method of claim 1 , wherein the one or more timeseries of synthetic aircraft performance data includes timeseries data emulating multiple flight parameters over each of one or more flights.
3 . The method of claim 2 , wherein the flight parameters include altitude, Mach number, total air temperature, ground speed, gross weight, fuel consumption, or a combination thereof.
4 . The method of claim 1 , wherein the one or more input parameters indicate how many of the timeseries of synthetic aircraft performance data are to be generated.
5 . The method of claim 1 , wherein the generative adversarial network corresponds to a generative adversarial network that is trained using one or more descriptors corresponding to: cruising altitude, initial weight, one or more weather conditions, an airspace, or a combination thereof.
6 . The method of claim 5 , wherein the one or more input parameters include: a number of flights, a particular cruising altitude, a particular initial weight, one or more particular weather conditions, a particular airspace, or a combination thereof.
7 . The method of claim 1 , wherein the processing of the training data to remove the outlier values and add the missing values includes:
receiving multiple timeseries of quick access recorded (QAR) data, each timeseries of the QAR data corresponding to one of the actual aircraft flights;
filtering parameters of the multiple timeseries of the QAR data based on one or more features of interest; and
processing the filtered parameters to remove the outlier values, add the missing values, or both, to generate multiple timeseries of processed input data.
8 . The method of claim 7 , wherein processing the filtered parameters further includes performing a smoothing operation and a normalization operation.
9 . The method of claim 7 , further comprising resampling each of the multiple timeseries of processed input data to have a uniform amount of time steps.
10 . The method of claim 1 , wherein:
the synthetic aircraft performance data generator includes:
a first layer of recurrent cells;
a second layer of recurrent cells, the second layer smaller than the first layer; and
a third layer of densely-connected cells; and
the discriminator includes:
a first convolution layer;
a second convolution layer;
a 1-dimension max-pooling layer;
a layer of recurrent cells; and
at least one layer of densely-connected cells.
11 . The method of claim 1 , further comprising storing the one or more timeseries of synthetic aircraft performance data for use in determining data analytics, aircraft performance analyses, or a combination thereof.
12 . A system comprising:
a memory configured to store instructions which, when executed by one or more processors, cause the one or more processors to:
obtain training data corresponding to multiple timeseries of multivariate aircraft performance data of actual aircraft flights;
process the training data to remove outlier values and add missing values;
perform a training operation of a generative adversarial network that includes a synthetic aircraft performance data generator and a discriminator, the training operation including:
updating the synthetic aircraft performance data generator to improve a likelihood that synthetic aircraft performance data generated by the synthetic aircraft performance data generator is identified as actual aircraft performance data by the discriminator; and
updating the discriminator to improve an ability of the discriminator to distinguish between the training data and the synthetic aircraft performance data generated by the synthetic aircraft performance data generator; and
after completion of the training operation:
receive one or more input parameters; and
generate, at the synthetic aircraft performance data generator, one or more timeseries of synthetic aircraft performance data based on the one or more input parameters.
13 . The system of claim 12 , wherein the one or more processors comprise the generative adversarial network including a synthetic aircraft performance data generator and a discriminator, wherein the generative adversarial network corresponds to a generative adversarial network that is trained using one or more descriptors corresponding to: cruising altitude, initial weight, one or more weather conditions, airspace identifier, or a combination thereof, and wherein the one or more input parameters include: a number of flights, a particular cruising altitude, a particular initial weight, one or more particular weather conditions, a particular airspace identifier, or a combination thereof.
14 . The system of claim 12 , wherein:
the synthetic aircraft performance data generator includes:
a first layer of recurrent cells;
a second layer of recurrent cells, the second layer smaller than the first layer; and
a third layer of densely-connected cells; and
the discriminator includes:
a first convolution layer;
a second convolution layer;
a 1-dimension max-pooling layer;
a layer of recurrent cells; and
at least one layer of densely-connected cells.
15 . The system of claim 12 , wherein the one or more input parameters indicate how many of the timeseries of synthetic aircraft performance data are to be generated.
16 . The system of claim 12 , wherein the instructions, when executed by the one or more processors, cause the one or more processors to process the training data to remove the outlier values and add the missing values including:
receiving multiple timeseries of quick access recorded (QAR) data, each timeseries of the QAR data corresponding to one of the actual aircraft flights;
filtering parameters of the multiple timeseries of the QAR data based on one or more features of interest; and
processing the filtered parameters to remove the outlier values, add the missing values, or both, to generate multiple timeseries of processed input data.
17 . The system of claim 16 , wherein the instructions, when executed by the one or more processors, cause the one or more processors to process the filtered parameters including performing a smoothing operation and a normalization operation.
18 . The system of claim 16 , wherein the instructions, when executed by the one or more processors, further cause the one or more processors to resample each of the multiple timeseries of processed input data to have a uniform amount of time steps.
19 . A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause the one or more processors to:
obtain training data corresponding to multiple timeseries of multivariate aircraft performance data of actual aircraft flights;
process the training data to remove outlier values and add missing values;
perform a training operation of a generative adversarial network that includes a synthetic aircraft performance data generator and a discriminator, the training operation including:
updating the synthetic aircraft performance data generator to improve a likelihood that synthetic aircraft performance data generated by the synthetic aircraft performance data generator is identified as actual aircraft performance data by the discriminator; and
updating the discriminator to improve an ability of the discriminator to distinguish between the training data and the synthetic aircraft performance data generated by the synthetic aircraft performance data generator; and
after completion of the training operation:
receive one or more input parameters; and
generate, at the synthetic aircraft performance data generator, one or more timeseries of synthetic aircraft performance data based on the one or more input parameters.
20 . The non-transitory computer-readable medium of claim 19 , wherein:
the synthetic aircraft performance data generator includes:
a first layer of recurrent cells;
a second layer of recurrent cells, the second layer smaller than the first layer; and
a third layer of densely-connected cells; and
the discriminator includes:
a first convolution layer;
a second convolution layer;
a 1-dimension max-pooling layer;
a layer of recurrent cells; and
at least one layer of densely-connected cells.